ReNikud: Audio-Supervised Hebrew Grapheme-to-Phoneme Conversion

2026-06-18

AI Summary

Modern Hebrew employs an abjad script that omits vowels, resulting in highly ambiguous grapheme-to-phoneme (G2P) conversion. Traditional approaches relying on diacritic-based vowel annotation (nikud) are constrained by data scarcity, failure to reflect spoken pronunciation, and absence of stress information. This work proposes ReNikud, the first method to leverage thousands of hours of unlabeled speech by generating phoneme-level ASR pseudo-labels to construct weak supervision signals. It introduces a pseudo-nikud architecture that directly predicts IPA phonemes at character positions and incorporates character–phoneme alignment as an inductive bias. By circumventing dependence on orthographic norms and manual annotation, ReNikud substantially outperforms prior state-of-the-art systems on both existing benchmarks and a newly curated, spoken-language-oriented MILIM benchmark, achieving more accurate modeling of everyday pronunciation.

Abstract

Grapheme-to-phoneme (G2P) conversion for Modern Hebrew is needed for applications like text-to-speech (TTS), but is challenging due to the language's abjad writing system, which leaves vowels largely unwritten, creating substantial ambiguity. Standard approaches first predict vowel diacritics (nikud) to produce International Phonetic Alphabet (IPA) transcriptions, but this is limited: vocalization data is scarce and laborious to produce, it does not specify features such as lexical stress, and it reflects formal grammatical rules rather than everyday spoken pronunciation. Direct sequence-to-sequence IPA prediction, meanwhile, struggles on limited data and fails to exploit the character-level alignment characteristic of abjads. Our method, ReNikud, overcomes these limitations with two key insights: (1) Weak audio supervision via a phoneme-based automatic speech recognition (ASR) pseudo-labeling pipeline on thousands of hours of unlabeled Hebrew audio, yielding phonemic transcriptions that reflect natural spoken norms without manual annotation. (2) A pseudo-vocalization architecture that predicts IPA phonemes at each character position, enforcing character-level alignment as an inductive bias. Results on existing Hebrew G2P benchmarks and the new targeted MILIM benchmark for spoken Hebrew show that ReNikud surpasses previous state-of-the-art methods. We will release our code and trained models to support further work on Hebrew TTS and speech technologies.

Problem

• Grapheme-to-phoneme
• Hebrew
• abjad
• vowel diacritics
• pronunciation ambiguity

Innovation

• grapheme-to-phoneme
• weak audio supervision
• pseudo-labeling
• character-level alignment
• Hebrew TTS